Room layout estimation obtaining method and system based on key point heat map correction

An acquisition method and acquisition system technology, which are applied in the field of room layout estimation and acquisition based on key point heat map correction, can solve the problems of the final performance of the model, the number of incorrectly connected areas, etc., to reduce learning difficulty, reduce overlapping areas, Convergence reduction effect

Active Publication Date: 2021-05-25

浙大宁波理工学院

8 Cites 0 Cited by

AI-Extracted Technical Summary

Problems solved by technology

[0002] At present, the key points of the key point heat map obtained by the neural network usually predict the key points that should exist at the boundary of the picture at a distance of tens of pixels from the boundary due to the prediction error of the network and the error of coor...

Method used

It should be noted that the correction boundary key point to the image boundary line of the key point heat map solves the pixel error problem caused by network prediction and coordinate scaling, and avoids the problem of miscalculating the number of connected areas of key points problem, greatly improving the accuracy of the model.

It should be noted that the correction of boundary key points to the image boundary line of the key point heat map solves the pixel error problem caused by network prediction and coordinate scaling, and avoids the problem of miscalculating the number of connected areas of key points problem, greatly improving the accuracy of the model.

It should be noted that, at first, after reordering the key points of the picture after flippi...

Abstract

The invention discloses a room layout estimation obtaining method and system based on key point heat map correction, and relates to the field of key point heat map correction, and the method comprises the steps: carrying out the overturning of a picture in a data set, carrying out the key point reordering, training a neural network model, obtaining a key point heat map and a room type of a to-be-recognized image through the neural network model, obtaining boundary key points of the key point heat map according to a corresponding relation between the key points of the room type and the key points of the key point heat map, obtaining image coordinates of the boundary key points, and correcting the boundary key points to an image boundary line of the key point heat map according to the image coordinates; finally, obtaining the connection relation of the corrected key point heat map according to the room type, and obtaining the room layout estimation according to the connection relation. According to the method, through key point reordering and boundary key point correction operation, the problem that the number of key point communication areas is wrong due to key point overlapping and pixel errors is solved, and the accuracy of the data set and the model is improved to a great extent.

Application Domain

Image enhancementImage analysis +2

Technology Topic

Computer visionNetwork model +7

Image

Examples

Experimental program(2)

Example Embodiment

[0036]Example one

[0037]Based on the key hot map acquired by the current neural network, we can understand that its boundary key is usually predicted in the position of dozens of pixels from the boundary, resulting in errors when the connection relationship is obtained according to the room type. The number of in-line regions has a great impact on the final performance of the model, followed by the data set used in training the neural network, and flip the data set in the data set, which can cause the data set picture. Data is not accurate, in order to solve these problems, improve the accuracy of models, such asfigure 1 As shown, the present invention proposes a room layout estimate acquisition method based on a critical point thermostat, which is implemented in accordance with the following steps:

[0038]The training method of neural network model, specifically includes the steps:

[0039]S01: Get the data set, the data set includes a key point mark map of multiple classes of preset rooms, by flipping the label diagram, and sequencing the extended data set after the flipped label graphs;

[0040]The method in which the sequencing in step S01 is: the order of the lanterns of the predetermid reference diagram is the ordering standard, so that the order of the launched labels is consistent with the order of the flipward reference.

[0041]It should be noted that first, after the key points of the flipped picture key points, each point is distributed relatively fixed in the picture, which reduces the difficulty of the neural network, so that the network can converge and start In order to reduce the effect of the error, second, the key point error is lowered, the calculation error of the pixel is improved, and the accuracy of the data set picture is improved.

[0042]In addition, the critical point overlap region is reduced, and the ambiguitability is reduced from the semantic to make the model forecast results have improved significantly.

[0043]S02: According to the extended data set training neural network model.

[0044]S1: Receive the image to be identified, acquire the key hot map of the image to be identified by the neural network model; the room type is composed of several ordered key points and contains several boundary critical points, between the key points Ordered connection;

[0045]S2: Get the boundary key point of the key hot map based on the key point of the room type and the key point of the key hot map;

[0046]It should be noted that the boundary key is the most peripheral point of the picture, seeimage 3 Point 2, 4, 6, 8,Figure 4Point 2, 3, 5, 6, thisimage 3 ,Figure 4The points are the key points of the room type. Each different type of room can be considered a cube box, and the vertex can be concatenated to the correct room layout estimate according to the specific order.

[0047]S3: Get the image coordinates of the boundary critical point and correct the boundary key to the image boundary line of the key hot map according to the image coordinate;

[0048]The correction method in the step S3 includes the steps of:

[0049]S31: Get vertical distance between the key points and the key thermostat boundary line according to the coordinates of each boundary;

[0050]S32: Get a key hot map boundary line with a minimum vertical distance and as a corrected boundary line;

[0051]S33: Move the boundary key to the correct boundary key to the correct boundary line in the vertical direction of the corrected boundary line.

[0052]S4: Obtain the connection relationship of the key hot map after the room type, and obtain the room layout estimate based on the connection relationship.

[0053]It should be noted that the correction boundary key to the image boundary line of the key hot map, solves the problem of pixel error due to the prediction and coordinate determination, while avoiding the problem of the number of wrong key communication regions, very It is highly enhanced the accuracy of the model.

Example Embodiment

[0054]Example 2

[0055]Such asfigure 2 As shown, the present invention proposes a room layout estimate acquisition system based on a key thermostat, including:

[0056]The training module of the neural network model, including:

[0057]Get the data set, the data set includes a key point of the multi-class preset room type, by flipping the label map and resets the extended data set after the flipped label graphs;

[0058]The rejected method is: in order of the lanterns of the preceding reference numeral diagram, the order of the launched labels is consistent with the order of the flipward reference.

[0059]It should be noted that first, after the key points of the flipped picture key points, each point is distributed relatively fixed in the picture, which reduces the difficulty of the neural network, so that the network can converge and start In order to reduce the effect of the error, second, the key point error is lowered, the calculation error of the pixel is improved, and the accuracy of the data set picture is improved.

[0060]In addition, the critical point overlap region is reduced, and the ambiguitability is reduced from the semantic to make the model forecast results have improved significantly.

[0061]According to the expanded data set training neural network model.

[0062]Neural network module, receives the image to be identified, acquire the key thermostat of the image to be identified by the neural network model; the type of room type is composed of several ordered key points, and contains several boundary key points, key points Orderly connection;

[0063]Boundary key module, obtain the boundary key of the key thermostat according to the key point of the key point of the key type of the critical point;

[0064]It should be noted that the boundary key is the most peripheral point of the picture, seeimage 3 Point 2, 4, 6, 8,Figure 4Point 2, 3, 5, 6, thisimage 3 ,Figure 4The points are the key points of the room type. Each different type of room can be considered a cube box, and the vertex can be concatenated to the correct room layout estimate according to the specific order.

[0065]The key hot map correction module, obtain the image coordinates of the boundary critical point and correct the boundary key to the image boundary line of the key hot map according to the image coordinate;

[0066]The correction method in the key hot map correction module includes:

[0067]The vertical distance of each boundary key and the key point heat map boundary line according to the coordinates of each boundary; obtain a key thermostat boundary line with the smallest vertical distance and the correction boundary line; the vertical direction of correcting the boundary line is the target direction Move boundary critical points to the corrected boundary line.

[0068]Boundary key connection module, obtain the connection relationship of the corrected key thermostat after the room type, and obtain the room layout estimate according to the connection relationship.

[0069]It should be noted that the correction boundary key to the image boundary line of the key hot map, solves the problem of pixel error due to the prediction and coordinate determination, while avoiding the problem of the number of wrong key communication regions, very It is highly enhanced the accuracy of the model.

PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.

Classification and recommendation of technical efficacy words

Reduce learning difficulty

Room layout estimation obtaining method and system based on key point heat map correction

AI-Extracted Technical Summary

Problems solved by technology

Method used

Abstract

Image

Examples

Example Embodiment

Example Embodiment

PUM

Description & Claims & Application Information

Similar technology patents

Quadruped robot motion control method based on reinforcement learning and position increment

License plate classification method, license plate classification device and computer readable storage medium

5-tone input method and its keyboard

Image processing method and device, electronic equipment and storage medium

Multi-functional flute

Classification and recommendation of technical efficacy words

5-tone input method and its keyboard

License plate classification method, license plate classification device and computer readable storage medium

Quadruped robot motion control method based on reinforcement learning and position increment

Multi-functional flute

Image processing method and device, electronic equipment and storage medium